Replit is no longer working with Anthropic, CEO says

Replit has cut ties with AI research lab Anthropic to run its cloud development environment on alternative model architectures, according to statements made by Replit chief executive officer and co-founder Amjad Masad onstage at TechCrunch Disrupt on Thursday. The decision marks an operational shift away from the Claude foundation models that previously powered Replit Agent and supported substantial revenue growth, as detailed in TechCrunch's report.

This report details the operational decisions, benchmarking data, contradictory partnership accounts, and software infrastructure considerations behind Replit's supplier changes for software engineers, startup founders, and enterprise technology leaders evaluating automated programming tools. The analysis examines executive claims regarding model performance and pricing, disputed vendor contracts, documented production deployments of Claude systems, and the enterprise requirements shaping autonomous agent deployments inside cloud development workspaces.

Replit is no longer working with Anthropic: the announcement

During an onstage appearance at TechCrunch Disrupt on Thursday, Replit chief executive officer Amjad Masad announced that his company is no longer working with Anthropic, citing cost, partnership terms, and model performance. Masad explained that ChatGPT is better now and noted that internal tracking mechanisms consistently favor alternative systems for building software applications from scratch. According to Masad, Replit publishes a public benchmark showing that models such as Sol and Astra consistently perform better during vibe-coded application development than competing options.

The executive announcement introduced immediate controversy because Replit had previously expanded under an alliance with Anthropic that powered the cloud workspace with Claude models. That technical collaboration helped Replit grow its commercial revenue significantly among developers. Despite Masad's onstage statements, an individual familiar with the corporate relationship between Anthropic and Replit rejected the claim, speaking on condition of anonymity to state that the partnership remains actively in place. The contradiction highlights industry friction surrounding model procurement agreements as developer platforms negotiate commercial licensing terms.

What actually changed with Replit is no longer working with Anthropic

The shift alters the underlying artificial intelligence foundation powering Replit Agent, transitioning browser-based application synthesis away from Anthropic's Claude ecosystem toward alternative infrastructure. While Replit announced a partnership with OpenAI in August, the company gave no public indication at that time that it planned to sever ties with Anthropic. Masad pointed directly to benchmark metrics, pricing, and software generation quality as decisive catalysts, explaining that models named Sol and Astra consistently surpassed competing options when constructing full-stack applications from scratch in browser sandboxes.

Despite declarations of an operational split, public records indicate that Anthropic technology remained active across Replit's production infrastructure in recent weeks. A technical blog post published by Replit in late September specifically identified Anthropic's Claude Fable 5 and Claude Fable 5.1 models as production engines running inside the live Replit Agent environment. Additionally, a Replit product employee posted on social network X around the same time praising the technical capabilities of Anthropic's flagship Opus 5.5 model during practical software engineering workflows.

Who should care about Replit is no longer working with Anthropic

Software developers, technical founders, and engineering teams relying on Replit to generate and deploy production software are directly affected by changes in foundation model providers. Replit sells a cloud development platform where users build and deploy applications directly from inside a web browser without configuring local machines. Because different foundation models exhibit varying code synthesis styles, token efficiency, and debugging capabilities, modifying the primary reasoning engine affects how conversational instructions translate into functional software. Developers must evaluate whether Sol and Astra produce application logic that matches previous Claude outputs.

Enterprise procurement leaders and technology directors must also track this transition as Replit rapidly expands its business footprint toward larger corporate clients. Masad noted onstage that while Replit is not disclosing annual recurring revenue totals, enterprise contracts grew from roughly two percent of revenue last year to fifteen percent this year. In addition, business-to-business transactions now generate more than fifty percent of the company's total revenue. As commercial clients increase their commitments to browser-based development environments, organizational buyers require clear visibility into model stability, vendor licensing, and compute costs.

How to try Replit is no longer working with Anthropic

Developers interested in evaluating Replit's current model infrastructure can access the cloud software platform directly through any standard web browser to build and launch software. The platform allows users to develop and deploy applications from scratch, using conversational prompting through Replit Agent to synthesize code without local machine installations. Following the partnership with OpenAI announced in August, developers can test the programming performance of Sol and Astra during end-to-end vibe-coding sessions. By comparing these builds against earlier Claude-generated projects, engineers can directly assess generative speed, algorithmic accuracy, and multi-file code scaffolding.

Evaluating these changes also involves understanding the environmental conditions Masad described as necessary for operating autonomous AI agents safely in production. Masad explained that just as human workers need comfortable offices, air-conditioning, computers, and food to work effectively, software agents require a dedicated habitat. According to Masad, that operational habitat must include effective orchestration, reliable error recovery, robust security sandboxing, dependable networking, and identity authorization. Developers testing the platform can observe how Replit enforces these sandboxed constraints when autonomous agents execute commands, manage project directories, resolve errors, and deploy functional applications.

What to watch after Replit is no longer working with Anthropic

The primary factor to monitor moving forward is whether Anthropic or Replit issues formal documentation clarifying the contradictory statements regarding their supplier agreement. While Masad stated publicly that Replit is no longer working with Anthropic due to cost and performance considerations, an insider familiar with the arrangement insists the collaboration remains in place. Furthermore, Replit's own technical publications and social media channels highlighted Claude Fable 5, Claude Fable 5.1, and Opus 5.5 running in production in late September. Industry observers are watching to see whether Replit completely phases out Claude models or maintains multi-vendor access options.

Industry analysts will also track Replit's financial and enterprise trajectory as corporate adoption continues to expand. With enterprise contracts rising from two percent of revenue last year to fifteen percent this year and B2B deals accounting for more than half of overall revenue, Replit faces greater scrutiny over infrastructure reliability and cost predictability. Observers will also track whether Replit's public benchmarks continue demonstrating performance advantages for alternative models like Sol and Astra over rival coding engines. As competition across artificial intelligence providers escalates, developer platforms may increasingly adjust foundation model partnerships to secure favorable commercial economics and compute access.